At the Hot Chips 34 annual high-performance computing technology conference, Intel demonstrated the computing power of its Xeon server processor, codenamed "Sapphire Rapids," paired with its server GPU, codenamed "Ponte Vecchio." The processors boasted 100 times the computing performance of NVIDIA's A2.5 GPU. Intel also previewed the Core processor designs, codenamed "Meteor Lake," "Arrow Lake," and "Lunar Lake," highlighting their use of Foveros connectivity technology and the UCIe specification to improve overall computing performance.
According to Intel, the server GPU codenamed "Ponte Vecchio" will have three basic configuration modes, including running with a Xeon server processor codenamed "Sapphire Rapids". It can also be used independently or connected through Intel oneAPI.
The 2-Stack Ponte Vecchio processor can deliver approximately 64 TFLOPS of computing power in FP32/FP52 mode, 32 TFLOPS of computing power in TF32 (XMX Float 419), 16 TFLOPS of computing power in BF16/FP839 mode, and 8 TFLOPS of computing power in INT1678 mode.
When executing miniBUDE computing loads, "Ponte Vecchio" will be able to exert twice the computing power of the A100 GPU, and when executing ExaSMR nuclear reaction simulations, "Ponte Vecchio" will also
As for after "Ponte Vecchio", Intel recently confirmed that it will continue to launch a server GPU code-named "Rialto Bridge", and it is expected to provide test samples to partners in mid-2023.
At Hot Chips 34, the annual high-performance computing technology conference, Intel also previewed the upcoming "Meteor Lake," "Arrow Lake," and "Lunar Lake" Core series processors. These will utilize Foveros interconnect technology to connect the CPU, GPU, I/O controllers, and various SoC components. The UCIe (Universal Chiplet Interconnect Express) specification accelerates interconnected computing efficiency, allowing computing components designed using different process technologies to operate collaboratively within the same processor, thereby improving overall computing performance.
Intel also explained that the Xeon D-2700 and 1700 series server processors are designed specifically for 5G, IoT, enterprise, and cloud applications. While meeting space constraints and low-power operation requirements, they can meet computing requirements such as 100G Ethernet, embedded encryption acceleration, time-coordinated computing (TCC), time-sensitive networking (TSN), and built-in AI processing optimization.
Intel also reiterated that FPGA technology remains a powerful and flexible hardware acceleration tool, especially promising in front-end radio frequency (RF) applications. Intel will integrate digital and analog chips, as well as chips from different process nodes and foundries, to achieve greater computing performance with shorter development time.
